Transpose ND HIP Kernel Optimizations #216
Merged
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
made changes in test suite to read read fixed input and comparision with golden output
made changes to update destination strides based on permute order
added golden input and outputs for 3D [2, 0, 1] transpose version removed golden inputs and outputs for 6D data
r-abishek
reviewed
Apr 30, 2024
src/modules/hip/kernel/transpose.hpp
Outdated
|
|
||
| int maxLength = dstStrides[0]; | ||
| int xAlignedLength = maxLength & ~7; // alignedLength for vectorized global loads | ||
| int xDiff = maxLength - xAlignedLength; // difference between roiWidth and alignedLength |
There was a problem hiding this comment.
Remove the xAlignedLength variable and just substitute with
int xDiff = maxLength - (maxLength & ~7); // difference between roiWidth and (alignedLength = maxLength & ~7)
r-abishek
reviewed
Apr 30, 2024
src/modules/cpu/kernel/transpose.hpp
Outdated
| *(srcPtr3 + 2 * srcGenericDescPtr->strides[1]), | ||
| *(srcPtr3 + 3 * srcGenericDescPtr->strides[1]), | ||
| *(srcPtr3 + 4 * srcGenericDescPtr->strides[1]), | ||
| *(srcPtr3 + 5 * srcGenericDescPtr->strides[1]), |
There was a problem hiding this comment.
This whole 1/2/3/4/5/6/7 * srcGenericDescPtr->strides[1] is constant.
Can be taken out of 4 loops!
|
Addressed all comments wrt to transpose HIP |
added 2d golden output for transpose
modified test suite code to display the transpose variant being tested
|
@r-abishek Please take another look if possible and let me know if further changes are needed |
r-abishek
reviewed
May 21, 2024
| for(; vectorLoopCount < alignedCols; vectorLoopCount += vectorIncrement) | ||
| { | ||
| __m256 pSrc[4]; | ||
| rpp_simd_load(rpp_load8_f32_to_f32_avx, srcPtrRow[0], &pSrc[0]); |
There was a problem hiding this comment.
@sampath1117 Pls add the AVX flags
r-abishek
reviewed
May 21, 2024
src/modules/cpu/kernel/transpose.hpp
Outdated
| dstPtr[i] += increment; | ||
| } | ||
|
|
||
| void rpp_store16_f32_f32_channelwise(Rpp32f **dstPtr, __m128 *p) |
There was a problem hiding this comment.
Can we make these helpers inline?
r-abishek
approved these changes
May 21, 2024
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
No description provided.